1 Bialystok Uni ver sity of Tech nol ogy, Fac ulty of Civil and En vi ron men tal En gi neer ing, Wiejska 45A, 15-351 Bia³ystok, Po land Sulewska, M., Zabielska-Adamska, K., 2015. Data com pres sion by prin ci pal com po nent anal y sis (PCA) in mod el ling of soil den sity pa ram e ters based on soil gran u la tion. Geo log i cal Quar terly, 59 (2): 400-407, doi: 10.7306/gq.1169The pa ram e ter for the den sity spec i fi ca tion of nat u rally com pacted non-co he sive soils and soils in em bank ments of hy draulic struc tures is the den sity in dex (I D ). The pa ram e ter used to con trol the qual ity of com pac tion of co he sive and non-co he sive soils ar ti fi cially thick ened, em bed ded in a va ri ety of em bank ments is the de gree of com pac tion (I S ). In or der to de ter mine the pa ram e ters of den sity (I D or I S ), com pac tion pa ram e ters (r dmin , r dmax or r ds , w opt ) should be ex am ined in a lab o ra tory, which often is a long and dif fi cult pro ce dure to carry out. There fore, there is a need for meth ods of im prov ing and short en ing the test of com pac tion pa ram e ters based on the de vel op ment and ap pli ca tion of use ful cor re la tions. Since com pac tion pa ram e ters are de pend ent on the soil gran u la tion, a method based on re gres sion and ar ti fi cial neu ral net works was ap plied to de velop required cor re la tions. Due to the large num ber of in put vari ables of neu ral net works in re la tion to the num ber of case stud ies, a PCA method was used to re duce the num ber of in put vari ables, which re sulted in re duc tion in the size of neu ral net works.Key words: ar ti fi cial neu ral net works, prin ci pal com po nent anal y sis, com pac tion pa ram e ters, min i mum and max i mum dry den sity of solid par ti cles, grain ing pa ram e ters.